I have this model which has an encoder-decoder structure. I want to change the “functional.interpolate” layer (which does upsampling layer) to “functional.max_unpool2d”. But as you can see in the following code in the “colab link” https://colab.research.google.com/drive/1UHjW5tEEG-ZwIEIlPzVuZSgvGnPuKVMI?usp=sharing, the max pooling is inside a Sequential, and I couldn’t get the indices of the tensor of max pooling layer. Are there any ideas on how to change the “functional.interpolate” layer to “functional.max_unpool2d” without changing a lot in the code. Really appreciate any help. Thank you so much in advance.
To use this layer in an
nn.Sequential container, you could take a look at this post, which shows some approaches how multiple inputs can be used.
Thank you so much @ptrblck. That is very helpful.
May I ask you, what is the advantage that can I expect from using “functional.max_unpool2d” instead of “functional.interpolate”?. What I have understood from my reading, that will yield a cheaper computation model. Is this correct?
I don’t know the theoretical complexities of both algorithms and how they are implemented.
If you are concerned about the performance of these layers, you could profile them via e.g. the